Summary
Kohei Yamaguchi is a Staff Field Application Engineer with 11 years of experience specializing in compiler and toolchain development for DNNs and AI accelerators. He has deep hands-on expertise in LLVM/MLIR (as an LLVM committer), C++, Rust, and Python, and has built compiler dialects and passes to optimize training and inference across accelerators like Graphcore and Tenstorrent. His background spans research-grade HPC and production deployments—porting proprietary LLMs to multiple runtimes, benchmarking with MLPerf, and operating a 100+ GPU cluster—so he understands both low-level optimization and real-world scale. Kohei prefers pure SWE roles focused on compilers and toolchains and brings a rare blend of research rigor and field engineering, having moved between R&D and commercial application at NTT and Tenstorrent. An interesting detail: he has directly integrated AI models into varied accelerator runtimes (llama.cpp, TensorRT-LLM, Poplar, TT-Buda/TT-Metal), demonstrating practical cross-platform portability.
11 years of coding experience
11 years of employment as a software developer
Ritsumeikan University
Japanese, English